Reducing costs through digitisation and condition based monitoring

This paper's focus is the advocation of utilising diagnostic data available from digital field devices to help reduce operating costs for end users. In recent years companies across multiple industrial sectors have invested in improving their understanding of both the historical and live data they produce. The source of the data is specific to the processes but the objective for all remains the same - to use statistical techniques to develop a toolset that can be used to predict performance based on live and historical data. For the oil and gas industry, the continued adoption of digital device transmitters has increased the volume of data available from instruments such as flow meters, temperature probes and pressure sensors. Typically, this additional data provides information on the integrity or quality of the associated device. However, with the appropriate level of facility and instrument knowledge it is also possible to infer information with respect to the process stream. Furthermore, this data, if correctly interpreted, can be used to predict maintenance and calibration requirements, resulting in reduced staff effort and shutdowns. The need for physical intervention due to device failure is also reduced, which in turn minimises the potential for accidental hydrocarbon release when a device is removed for repair or replacement. NEL are currently undertaking research projects with the primary objective of developing definitive correlations between process effects, meter condition and diagnostic data response. The paper provides details of said research, with particular reference to the data science and mathematical techniques currently being trialed for the analysis stage. The techniques, when fully developed, will be metering technology specific and therefore offer a level of insight to end users on facility and meter performance which is not currently available in industry. The toolsets developed will in turn provide the end users with the knowledge and confidence to make cost saving decisions with respect to planned maintenance as well as improving facility efficiency through a more comprehensive understanding of their own data sets. In recent years there has been an increase in the uptake of digital instrumentation that make use of fieldbus networks. Such devices are capable of outputting vast quantities of data sets over multiple digital protocols such as networks such as Modbus, Foundation Fieldbus, Profibus and more recently WirelessHart [1]. For the purposes of this paper, the data contained within these ‘big data’ sets can be split into two categories of process values, primary and secondary. The primary values relate to the device's principle function such as mass flow rate, density or pressure measurement. The secondary values relate to individual parameters that while available to the end user via fieldbus, require device specific knowledge to interpret. Historically, this secondary data has been used either internally within the instrumentation firmware to make automatic corrections or by metering technicians and commissioning engineers with specific knowledge of device manufacture and operations. As end user experience with these types of devices grows, there is a corresponding increase in a willingness to log these additional process values to facilitate more efficient operations and therefore reduce costs. There are at present several software packages that make use of machine learning statistical modelling to predict device performance over extended periods of time by identifying correlations between the secondary process values and the larger facility instrumentation installed as part of a given process. A specific application for this data logging paradigm is adopting ‘condition-based calibration’ (CBC) over the traditional ‘time-based calibration’ (TBC) With the TBC method there is potential to incur unnecessary costs through stopping operations/ productions in order to calibrate a flow meter, which has not deviated from its previous calibration curve. The total costs when one considers the associated operational tasks such as pipe fitting, electrical isolation/ connection, facility down time, meter shipping and calibration lab fees can be in the region of £30,000. A different but no less severe disadvantage of TBC is the scenario where a meter has significantly drifted from its last calibration but is not yet due for recalibration, resulting in measurement errors that in turn will have financial consequences for the end user that will not be picked up until the meters next scheduled calibration. A CBC schedule can address these challenges by allowing facilities to adopt a dynamic operating patterns. Such a system is based on autonomous data analysis of both primary and secondary process value performance over time. Only when a device such as a flow meter is shown to have drifted beyond is normal operating conditions based on the historical data held by the CBC system is the user alerted that action is required. Ideally this action is determined by the specifics of data output by the system. For example, the system should be able to distinguish between device performance degradation due to damage or erosion and a change in process conditions, which has affected meter performance by taking it outwith recommended operating limits.